[Programcode]

Why

src/lib/ai/client.ts was 1,585 lines with every AI feature in one file. The server side was split per feature in #75. This does the same for the client, as item 11 of plans/2026-10-01-next-15.md.

What changed

• The shared code moves to four files next to client.ts. request.ts holds the one transport, callSupabaseAI. errors.ts holds the error classes, sources.ts the discovered-source parsing and autosave, and text.ts the string helpers.
• Each feature gets its own file in src/lib/ai/features/, with the same seven names the server uses.
• client.ts now only re-exports. It lists the same 39 names, so no caller or test changes. A lint rule stops app code from importing the files behind it.

Scope

• This is a move only. Every function body is byte-identical to the original. The one in-body edit adds export to 11 helpers that other files now need.
• Two known smells stay as they are because fixing them would change behavior. Sparring validates its input twice, and callSupabaseAI has an unreachable TimeoutError branch.
• session-store.ts is out of scope, as the plan says.

Tradeoffs

streaming.ts and error-display.ts still import from ./client, not from the new files. Three page tests mock the whole of @/lib/ai/client. If error-display.ts imported errors.ts directly, those tests would get the real error class instead of the mock.

Blast Radius

The split touches every AI feature in the browser, but only the module layout changes. If something is wrong, it shows up as a broken import at build time or a failed AI request in the app. The payload check below rules out the second for every flow the verify recipes drive.

Verification

• Recorded every request body the fake model provider receives across verify-debatelab drive all. Two runs on main and one run on this branch give the same 140 requests, byte-identical after masking UUIDs, timestamps and HMAC signatures: sha256 ca3e696c… all three times. As a check that the comparison can fail, a one-word change to the client's default factoids region changed the recorded payload while its recipe still passed.
• drive all: 66 passed, 4 skipped on both main and this branch, desktop and mobile Chromium, fake provider.
• npm test (31 files, 280 tests, no test file changed), typecheck, lint, build and docs:check all pass locally. A partition script confirms each of the 79 original declarations appears exactly once with an unchanged body.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Reliability Improvements
  • AI-generated responses are more consistently checked and cleaned before being shown, helping prevent malformed results.
  • AI errors are handled more consistently, including clearer handling of timeouts, access limits, and provider failures.
  • Ask requests now have safeguards for oversized messages and conversation history.
• Source Handling
  • Discovered sources can be saved to the relevant debate or workspace, and factoid results include only verified citations.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
debate-navigator	Ready	Preview	Oct 16, 2026 2:40pm UTC
debatelab-staging	Ready	Preview	Oct 16, 2026 2:40pm UTC

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[coderabbitai]

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Repository YAML (base), Organization UI (inherited)
• Review profile: CHILL
• Plan: Free
• Run ID: 49dd8e25-897f-41e4-a2a9-75a13622567e

📥 Commits

Reviewing files that changed from the base of the PR and between 5af8a46 and 2cec375.

📒 Files selected for processing: 17

• AGENTS.md
• docs/architecture.md
• docs/standards/boundaries.md
• e2e/fixtures/ai-responses.ts
• eslint.config.js
• src/lib/ai/client.ts
• src/lib/ai/errors.ts
• src/lib/ai/features/ask.ts
• src/lib/ai/features/attack.ts
• src/lib/ai/features/brainstorm.ts
• src/lib/ai/features/crossfire.ts
• src/lib/ai/features/factoids.ts
• src/lib/ai/features/generate.ts
• src/lib/ai/features/sparring.ts
• src/lib/ai/request.ts
• src/lib/ai/sources.ts
• src/lib/ai/text.ts

Included review availability: This review used your included allowance. Your plan provides up to 1 included review per hour; 0 remain after this review.

📝 Walkthrough

The AI client implementation is divided into shared request, error, source, and text modules and feature-specific modules. The client facade re-exports those APIs. ESLint rules and documentation describe the internal import boundary and module layout.

Changes: AI Client and Feature APIs

Layer / File(s)	Summary
Shared AI errors, sources, and text utilities src/lib/ai/errors.ts, src/lib/ai/sources.ts, src/lib/ai/text.ts	Added typed AI errors and classification, source parsing and autosaving, and UTF-8 byte and generated-output utilities.
AI request transport src/lib/ai/request.ts	Added authenticated request handling, context and abort handling, error reporting, body-size limits, and Supabase error-payload parsing.
Ask and generated-content features src/lib/ai/features/ask.ts, src/lib/ai/features/attack.ts, src/lib/ai/features/brainstorm.ts, src/lib/ai/features/factoids.ts, src/lib/ai/features/generate.ts	Added Ask input bounds and response validation, plus attack, brainstorm, factoids, and case-generation operations with feature-specific response processing.
Crossfire and sparring operations src/lib/ai/features/crossfire.ts, src/lib/ai/features/sparring.ts	Added Crossfire turn, debrief, and signal requests, and sparring turn generation and session judging with input and response validation.
Client facade and import boundary src/lib/ai/client.ts, eslint.config.js, AGENTS.md, docs/architecture.md, docs/standards/boundaries.md, e2e/fixtures/ai-responses.ts	The client module now re-exports APIs from their modules. ESLint restricts direct imports of internal AI modules. Documentation and a fixture comment reference the updated module layout.

Estimated code review effort: 4 (Complex) | ~60 minutes

Sequence Diagram(s)

sequenceDiagram
  participant generateCase
  participant callSupabaseAI
  participant SupabaseAIFunction
  participant autosaveDiscoveredSources
  generateCase->>callSupabaseAI: submit feature input
  callSupabaseAI->>SupabaseAIFunction: send authenticated AI request
  SupabaseAIFunction-->>callSupabaseAI: return AI response
  callSupabaseAI-->>generateCase: return result and discovered sources
  generateCase->>autosaveDiscoveredSources: save parsed discovered sources

Loading

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Note

🎁 Summarized by CodeRabbit Free

Your organization is on the Free plan. CodeRabbit will generate a high-level summary and a walkthrough for each pull request. For a comprehensive line-by-line review, please upgrade your subscription to CodeRabbit Essentials by visiting https://app.coderabbit.ai/settings/billing.

_{Comment @coderabbitai help to get the list of available commands.}

[Programcode]

GitHub Actions did not run on this PR. Every job stopped after 2 seconds with "recent account payments have failed". These are the local results at head 2cec375f. They are local passes, not CI passes.

• npm run verify: exit 0.
• ./.superset/db-gate.sh: exit 1. 65 passed, 6 failed, 1 skipped.
• The same gate on main (5af8a46e) gives the same result: the same 6 tests fail, 65 pass, 1 is skipped.

All 6 failures are the same 3 specs on desktop and mobile Chromium: e2e/full-stack/case-generation, current-debate and debate-context. Each one waits for a Library textbox named "Source title...". Since commit 2dc73ff (design audit, shell) the fields are labelled "Title" and "Content". That audit merged while Actions was not running jobs. This PR touches neither the Library page nor those specs. The fix belongs in its own PR.

[Programcode]

Codex review (gpt-6.1-sol, high) of head 2cec375: no actionable findings. Two earlier internal rounds (Opus and GPT) found gaps in the lint guard: specifiers with extensions, dynamic import(), and false positives on the server's modules. Commit 2cec375 fixes them.